What about Underevaluating Operational Value at Risk in the Banking Sector?

Abstract

The objective of this article is to develop a precise and rigorous measurement of a bank's operational VaR. We compare our model to the standard model frequency used in practice. This standard model is constructed based on lognormal and Poisson distributions which do not take into account any data which fall below the truncature threshold and undervalue bank's exposure to risk. Our risk measurement also brings into account external operational losses that have been scaled to the studied bank. This, in effect, allows us to account for certain possible extreme losses which have not yet occurred. The GB2 proves to be a good candidate for consideration when determining the severity distribution of operational losses. As the GB2 has already been applied recently in several financial domains, this article argues in favor of the relevance of its application in modeling operational risk. For the tails of the distributions, we have chosen the Pareto distribution. We have also shown that the Poisson model, unlike the negative-binomial model, is retained in none of the cases for frequencies. Finally, we show that the operational VaR is largely underestimated when the calculations are based solely on internal data.